Projects
Workshops
- BMVC 2022: Universal Representations
Co-organised with Wei-Hong Li, Hakan Bilen, Ales Leonardis, Xialei Liu, and Steven McDonagh. - ECCV 2022: Sign Language Recognition, Translation & Production
Co-organised with Liliane Momeni, Gül Varol, Hannah Bull, Prajwal KR, Ben Saunders, Neil Fox, Necati Cihan Camgöz, Richard Bowden, Andrew Zisserman and Bencie Wollr. - NeurIPS 2021: The pre-registration experiment: an alternative publication model for machine learning research (second edition)
Co-organised with Luca Bertinetto, João F. Henriques, Alex Hernández-García, Hazel Doughty, and Gül Varol. Video of the event can be found here. Slides for the introductory remarks can be found here. The PMLR proceedings from the workshop can be found here. - NeurIPS 2020: The pre-registration experiment: an alternative publication model for machine learning research
Co-organised with Luca Bertinetto, João F. Henriques, Michela Paganini and Gül Varol. The PMLR proceedings from the workshop can be found here. - ECCV 2020: Sign Language Recognition, Translation & Production
Co-organised with Necati Cihan Camgöz, Gül Varol Neil Fox, Richard Bowden, Andrew Zisserman and Kearsy Cormier. - CVPR 2020: The End-of-End-to-End: A Video Understanding Pentathlon
Co-organised with Yang Liu. Arsha Nagrani Antoine Miech, Ernesto Coto, Ivan Laptev, Rahul Sukthankar, Bernard Ghanem and Andrew Zisserman. - ICCV 2019: Neural Architects: What have we learned and where are we going?
slides and papers, wonderful reviewers,
Co-organised with Li Shen, Jie Hu, Barret Zoph, Andrea Vedaldi and Andrew Zisserman. Supported by Momenta. - ICCV 2019: Should we preregister experiments in computer vision?
Co-organised with João F. Henriques, Luca Bertinetto and Jack Valmadre. Supported by Unitary
PhD Examiner
-
10 / 2022
Bo Zhao (University of Edinburgh, examined with Iain Murray)
ML Competitions
- Condensed Movies Challenge 2021, 1st place with Ioana Croitoru, Simion-Vlad Bogolin, Marius Leordeanu, Hailin Jin, Andrew Zisserman and Yang Liu.
- Multi-Moments-in-Time Challenge 2019, challenge report, leaderboard, 3rd place with Yang Liu, Qingchao Chen and Andrew Zisserman
- LSMDC Challenge 2019 leaderboard 2nd place with Andrew Brown, Yang Liu, Arsha Nagrani and Andrew Zisserman
Reviewing
- conferences: CVPR, ICCV, ECCV, NeurIPS, ICML, BMVC, AAAI, SIGGRAPH
- journals: TPAMI, IJCV, TIFS, TMM
- area chair: BMVC 2021, WACV 2023
- outstanding reviewer awards: CVPR 2019, 2020, 2021, ICCV 2021, RC 2022
- Turing institute: PDEA 2021
- Workshops: NeurIPS Reproducibility Challenge 2019, CVPR'20 Workshop on Scalability in Autonomous Driving, CVPR 2020 Workshop on Neural Architecture Search and Beyond for Representation Learning, Seventh International Workshop on Egocentric Perception, Interaction and Computing, ICCV2021: 4th Workshop on Closing the loop between Vision and Language, AI for Creative Video Editing and Understanding 2021, ML Reproducibility Challenge Fall 2021
Misc
- Convnet burden - A tool for estimating the memory/computational costs associated with neural network architectures.
- Yaspi - Yet Another Slurm Python Interface.
- slurm_gpustat - Command line summaries of GPU cluster utilization.
- mcnPyTorch/pytorch-mcn - Tools for converting models between the MatConvNet and PyTorch frameworks.
- Object detector implementions - mcnSSD/mcnFasterRCNN/mcnRFCN.
- Computing Euclidean Distance Matrices: Euclidean Distance Matrix Trick.
- Pixurgery - An open source iPhone app for applying facial effects to live video. Built with code gurus Tom and Tom (code here). YouTube coverage by Youtube celeb GoGoManTV. 1
- A few notes on some topics in Computer Vision: Differentials for Applied Computer Vision.
- Some proper engineering, but not by me.
Talks
-
11 / 2022
Invited talk, AIMS, University of Oxford -
10 / 2022
Invited seminar, Observatory for Human Machine Collaboration, Cambridge -
10 / 2022
WASP visit, Cambridge University -
11 / 2021
Machine Learning Workshop, Sirius University, Sochi, Russia (Virtual) -
06 / 2021
CVPR 2021 mentoring session (slides), (Virtual) -
03 / 2021
Queen's University, Kingston, Canada (Virtual) -
02 / 2021
Explainable Machine Learning, University of Tübingen, Tübingen, Germany (Virtual) -
12 / 2020
Perceiving Systems, Max Planck Institute for Intelligent Systems, Tübingen, Germany (Virtual) -
12 / 2020
DAADNet-tour mini seminars, Germany (Virtual) -
09 / 2020
Visual Geometry Group, University of Oxford, UK -
07 / 2020
Sydney ML Meetup, Sydney, Australia (Virtual) -
09 / 2019
Google Research, New York, USA (Virtual) -
03 / 2019
Mathematics Department, University of Oxford, UK
Serious research
- Stale Diffusion - SIGBOVIK 2024. I may not have contributed to this paper.
- Large Language Models are Few-shot Publication Scoopers - SIGBOVIK 2023. Contrived with Lili Momeni and João F. Henriques
[paper] [code] [video] [slides] - A 23 MW data centre is all you need - SIGBOVIK 2022. Forecast with Dylan Campbell and João F. Henriques
[paper] [code] [video] [slides] - On the origin of species of self-supervised learning - SIGBOVIK 2021. Evolved with Erika Lu and João F. Henriques
[paper] [video] [slides] - State-of-Art-Reviewing: A Radical Proposal to Improve Scientific Publication - SIGBOVIK 2020. This paper was written with Jamie Thewmore, Robert McCraith and João F. Henriques.
[paper] [video (review by Yannic Kilcher)] - Deep Industrial Espionage - Narrowly missed the deadline for SIGBOVIK 2019. This paper was written with James Thewlis, Sébastien Ehrhardt and João F. Henriques. This paper was awarded Most timely paper at SIGBOVIK 2020.
- Substitute Teacher Networks: Learning with Almost No Supervision - Received the following comment by Reviewer 2 at SIGBOVIK 2018: "The paper makes important advances in the area...". This paper was handsewn with James Thewlis, and João F. Henriques. Code to run one of the experiments described in the paper can be found here.
- Stopping GAN violence: Generative Unadversarial Networks - Winner of the award for excellence in coincidentally/blindly writing, creating titles and submitting similarly/ relatedly/with related titles at SIGBOVIK 2017. This paper was co-manufactured with Sébastien Ehrhardt and João F. Henriques. Code to reproduce the experiments described in the paper can be found here.
Ye Olden Times
In a previous life, I enjoyed rowing.
Art by others
Unsolicited links to beautiful creations.
- Drawings and painting by Joana d'Arte
- Creative writing by David Fouhey and Daniel Maturana
- Paintings by Hannah Bull
CV
Personal
About six foot tall, in the morning. Star sign: Cancer, so I’m sensitive to the feelings of others and have a vivid imagination, but I can also seem moody in stressful situations, especially around a Sagittarius. My shoe size is 11 UK (or 13 US), but my feet are quite wide, so I struggle to find shoes in standard retailers such as Clarks.
Academic History
I was once described as “quite clever” by my tutor, and at the precocious age of just 20, almost beat my younger brother at a game of chess. He is however, by his own admission, not very good.
By the age of 14, I could read. In English, mainly, but also in a language I made up in my head. It had only one vowel, o, so difficult pronunciation was also a skill I got to grips with early.
Employment History
Shortly after my voice had dropped, I was out on the farmland as a beater in the local shoot. I’m widely feared by pheasants and foxes near Shelfanger, a metropolis of commerce and culture in Norfolk (UK). From these experiences, I have developed the skill of using loud shouting into nearby woodland to resolve confrontation.
As soon as I could taste the difference between Adnams Southworld Bitter and Adnams Broadside, I embarked on my next major career move, spending three and a half weeks working on a worm farm (selecting the juiciest and finest worms to be used as bait). This allowed me to develop an appreciation for quality, technical agricultural practice, and the colour of a healthy worm.
My next opportunity for gainful employment arose in the West Indies at an insurance consultancy firm, where I developed the ability to pour a glass of rum under pressure. I followed this up with a trip to China, where I taught literally tens of children the British spellings of words in the English language. They called me “Shan Mu” (mountain berry). This name arose from an alcohol fuelled argument over the definition of a berry, in a bar in down town Shanghai. The observant linguist will notice the phonetic similarity to my forename, Samuel.
I followed this position with a job in a steel factory, thirty seven minutes and fifteen seconds (on a road bike, into a mild headwind, no precipitation, mild cloud cover) away from my home. Any prospective employer should therefore note that I am comfortable with the harsh realities of commuting. Here, I learned how to keep up the appearance of work in front of factory supervisors, whilst contributing the A.M (absolute minimum), towards efficiency and production. I also learned to develop my considerable linguistic skills, and can almost count to seventeen fluently in Polish.
Other Interests And Achievements
Most people struggle to name their proudest achievement. As this CV shows, if you are looking for a member of the “most” category, I’m not the candidate for you. Proudest achievement: My Yahtzee record against my Grandmother:
Wins 14, Losses 3.
I’m also a keen sportsman. I played as the second ranked player in the “Attleborough Allstars” B team, in the Norfolk Table Tennis Association, Third Division. At my final school sportsday, I came second in the 800m. Honesty and integrity, are however, also qualities that come easily to me, and it is worth noting that several high performing athletes were ill at the time, having contracted food poisoning the night before at my birthday party. I cooked the chicken.
Poetry
The following piece was inspired by real life experiences and was rejected from a regional poetry competition in 2016 1:
Computer you piece of @#$%&!
Now is not the time.
The deadline approaches.
Other Employment
Paper round, stretching from house 34, Lewis Lane, Diss (reads the Telegraph), to 47, Appletetre Lane, Diss (reads the sun, has a Doberman called Spike).
Worked as a tyre consultant in a tyre garage in Diss for three months, before legal complications arose regarding age of employment.
Referees
- Howard Webb I’ve always felt that he was a little quick to hand out a first yellow, but his firm but fair approach to diving in and around the penalty area won him respect amongst fans and club management the world over.
- Sergio Pezzotta Never afraid of controversy, or intimidated by the big names in the dug outs, his favourite colour was red. He tended to feel that eleven men on each team made the beautiful game a little too crowded.
- Pierluigi Collina One of the games greatest. In addition to an abundance of wisdom and great people skills, his prominent eyeballs gave him a physiological advantage over his colleagues, leaving many to question whether he was entirely human.
1. Robert Pinksy has a CV three pages long. It does not contain a single poem. The former Poet Laureate of the United States does have a section entitled Poetry. However, rather than delighting the reader with his poems right then and there “mid-CV” and sacrificing valuable page space for his art, he simply lists them by name, together with the names of the numerous rewards that they have received. In contrast to the poetic success enjoyed by RP, my poem was rejected. The editor did helpfully offer to give me feedback on mywork - an offer which I declined physically. Poetry is not about compromise (NOTE: this statement is not an absolute - I removed a key line of the poem to ensure that it fitted into a tweet).↩